A Genetic Algorithm for Multi-Capacity Fixed-Charge Flow Network Design

Caleb Eardley,Dalton Gomez,Ryan Dupuis,Michael Papadopoulos,Sean Yaw

2024-10-26

Abstract:The Multi-Capacity Fixed-Charge Network Flow (MC-FCNF) problem, a generalization of the Fixed-Charge Network Flow problem, aims to assign capacities to edges in a flow network such that a target amount of flow can be hosted at minimum cost. The cost model for both problems dictates that the fixed cost of an edge is incurred for any non-zero amount of flow hosted by that edge. This problem naturally arises in many areas including infrastructure design, transportation, telecommunications, and supply chain management. The MC-FCNF problem is NP-Hard, so solving large instances using exact techniques is impractical. This paper presents a genetic algorithm designed to quickly find high-quality flow solutions to the MC-FCNF problem. The genetic algorithm uses a novel solution representation scheme that eliminates the need to repair invalid flow solutions, which is an issue common to many other genetic algorithms for the MC-FCNF problem. The genetic algorithm's efficiency is displayed with an evaluation using real-world CO2 capture and storage infrastructure design data. The evaluation results highlight the genetic algorithm's potential for solving large-scale network design problems.

Neural and Evolutionary Computing,Artificial Intelligence

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper aims to solve the multi - capacity fixed - cost network flow (MC - FCNF) problem. Specifically, the MC - FCNF problem is a generalization of the fixed - cost network flow (FCNF) problem. Its goal is to assign capacity to each edge in the network to carry the specified target traffic at the minimum cost. Different from FCNF, MC - FCNF allows each edge to have multiple capacity options, and each capacity has a corresponding fixed construction cost and variable usage cost.

#### Description of the main problem

1. **Problem background**:
- The MC - FCNF problem naturally arises in fields such as infrastructure design, transportation, telecommunications, and supply chain management.
- This problem belongs to the NP - Hard category, so for large - scale instances, using exact solution methods (such as integer linear programming) is impractical.

2. **Limitations of existing solutions**:
- Existing genetic algorithms usually need to repair invalid flow solutions when dealing with the MC - FCNF problem, which increases the computational complexity.
- Traditional genetic algorithm representation methods (such as using binary variables to represent edge usage or predefined paths) may lead to sub - optimal solutions or require complex repair operations.

3. **Solution proposed in the paper**:
- The paper proposes a new genetic algorithm. This algorithm avoids the need to repair invalid solutions by introducing a fixed - cost scaling parameter (dek) and representing each solution as an array of parameters.
- This representation method ensures that each solution is a valid flow allocation scheme, thus improving the efficiency of the algorithm.
- This genetic algorithm also combines linear programming (LP) techniques to quickly find high - quality approximate solutions.

4. **Application examples**:
- The paper is evaluated by using real - world carbon dioxide capture and storage (CCS) infrastructure design data, showing the potential of this algorithm in large - scale network design problems.
- The evaluation results show that this genetic algorithm performs well when dealing with very large network instances. Even compared with the optimal integer linear programming (ILP) solution, it can find a near - optimal solution in a relatively short time.

### Summary

The main contribution of this paper is to propose a novel genetic algorithm to efficiently solve the MC - FCNF problem. In particular, by introducing a fixed - cost scaling parameter to simplify the representation and optimization process, it avoids the problem of repairing invalid solutions common in traditional methods. This improvement enables the algorithm to handle large - scale instances more effectively and shows good performance in practical applications.